Introduction

For many students, undertaking a PhD research program can be an immensely stressful period in their lives. This is evidenced by studies reporting increased rates of mental health concerns or problems in PhD students across the globe1,2,3. A United Kingdom (UK)-based study by Hazell et al.3 reported that a large proportion of PhD students experience mental health concerns. Approximately 70% of PhD students in their sample self-reported clinically relevant depression symptoms (i.e., mild to severe symptoms), 75% self-reported clinically relevant anxiety symptoms, and 35% self-reported being at risk of suicide. Critically, these mental health symptoms were not explained by mental health concerns that existed before commencing their PhD studies3. Similarly, an Australian-based study by Brownlow et al.1, reported that Higher Degree Research (HDR) students (i.e., Master of Research or PhD students) had a greater risk of experiencing mental health concerns compared to the general population during COVID-19 lockdowns. Approximately 25% of HDR students were at risk of experiencing mental health concerns compared to 15.7% of the general community1.

Research on the mental health of university students is well-established. However, studies within this area often consider university students as a broad population, and thus, analyses conducted often do not disaggregate the various student sub-populations. For example, undergraduate student populations are not considered separately from postgraduate student populations, or PhD student populations from other postgraduate student populations. Disaggregation is important due to unique mental health concerns and challenges within each student sub-population. Specifically, PhD students face unique challenges such as time constraints, a lack of a sense of belonging, and uncertainty about the PhD process4. These challenges can result in poor mental health in PhD students, which can not only impact their study progress5,6 but also result in broader adverse effects on their general functioning post-PhD study7. However, the paucity of research that focuses specifically on the mental health of PhD students means that our knowledge base on the prevalence and determinants of mental health in this population remains limited.

Recently, Berry et al.8 sought to address this limited understanding by investigating the predictors of mental health symptoms in a sample of PhD students in the UK. Via a large-scale nationwide assessment in 2018ā€“2019 (Understanding DOCtoral researcher mental health; U-DOC study), they investigated the demographic, occupational, psychological, social, and relational determinants of three mental health outcomes: depression, anxiety, and suicidality. They reported that psychological, social, and relational determinants were the strongest and most consistent in predicting depression, anxiety, and suicidality in this population. Namely, these were the psychological determinants of more imposter thoughts and higher levels of perfectionism, the social determinant of greater loneliness, and the relational determinant of reduced supervisory relationship communion (i.e., proximity and cooperativeness between student and supervisor). The identified determinants of mental health in this population represent risk factors that are common in other university student populations and the general community (e.g., loneliness9,10), but also factors that are distinct to the PhD student population (e.g., supervisory relationship11). Findings from the study by Berry et al.8 highlight the need for analyses in future research studies to disaggregate PhD students from other student populations and more importantly, contribute to our understanding of the mental health of PhD students.

Research on the mental health of PhD students has often focused on student populations in the UK and the United States of America12. Comparatively, research on the mental health of PhD students in Australia is lacking12, representing a significant research population gap. We cannot assume that the determinants of the mental health of PhD students in the UK8, or other countries12, would be identical to those of the mental health of PhD students in Australia, especially since PhD program structures, demands, and requirements vary worldwide13. For example, the oral examination is widely implemented as part of the assessment of the PhD in the UK but not in Australia. Thus, a study focused on the PhD student population in Australia is needed, to not only contribute to the broader research field by gathering more research evidence on this topic but also inform the Australian higher education system on the potential strategies to improve the mental health of PhD students in Australia.

The present study was motivated by the UK-based study by Berry et al.8. To extend their findings and to address the aforementioned research population gap, we aimed to identify the determinants of the mental health of PhD students in Australia. Similar to the study by Berry et al.8, the present study focused on identifying the demographic, occupational, psychological, social, and relational determinants of three mental health outcomes: depression, anxiety, and suicidality. Given the differences in PhD programs across the UK and Australia, there were difficulties generalising existing literature to our sample population, and therefore no a priori hypotheses were formulated.

Methods

The present study forms part of a broader research project. Only data from measures relevant to the specific aim of this study are analysed and reported in this article. Data from other measures will be analysed and reported elsewhere.

Participant recruitment

Individuals undertaking a PhD program at an Australian university, aged 21Ā years or older, and residing in Australia were invited to participate in an online survey. Individuals were recruited through social media advertisements (e.g., Twitter/X, LinkedIn) and third-party emails to contacts at relevant institutions. The online survey was open from November 2022 to April 2023. Although participants did not receive financial compensation or incentive for their participation, they were informed that a donation of AUD$1.00 per survey response (up to a total of AUD$300.00) would be made to an Australian mental health organisation. The study was approved by the University of the Sunshine Coastā€™s Human Research Ethics Committee (Approval number: S221753) and was performed in accordance with the human research ethics guidelines.

A total of 389 PhD students participated in the study. Data from 87 participants were excluded from the analysis due to their withdrawal from the study (in line with the approved ethics requirements). This resulted in a final sample of 302 PhD students included in the analysis.

Measures

Mental health symptoms

Depression

The 9-item Patient Health Questionnaire (PHQ-9)14 was used to measure depression severity. The PHQ-9 asks respondents to consider their feelings over the last two weeks and indicate the frequency with which they have experienced depressive symptoms (response items) on a 4-point Likert scale (0 ā€˜Not at allā€™ to 3 ā€˜Nearly every dayā€™). Scores on each item are summed. The total score can range from zero to 27, with higher scores indicating greater severity of depressive symptoms. The PHQ-9 had good reliability in the current sample (Ī±ā€‰=ā€‰0.890)15.

Anxiety

The 7-item Generalised Anxiety Disorder (GAD-7) Assessment16 was used to measure anxiety. The GAD-7 asks respondents to consider their feelings over the last two weeks and indicate the frequency with which they have experienced symptoms of anxiety (response items) on a 4-point Likert scale (0 ā€˜Not at allā€™ to 3 ā€˜Nearly every dayā€™). Scores on each item are summed. The total score can range from zero to 21, with higher scores indicating greater severity of anxiety symptoms. The GAD-7 had excellent reliability in the current sample (Ī±ā€‰=ā€‰0.911)15.

Suicidality

The 4-item Suicide Behavior Questionnaire-Revised (SBQ-R)17 was used to measure suicidality and suicide risk. Each item in the SBQ-R measures a different dimension of suicidality: lifetime suicide ideation and suicide attempt, frequency of suicidal ideation over the past 12Ā months, threat of suicidal behaviour, and likelihood of suicidal behaviour. The items differed in their response options17, but scores on each item are summed. The total score can range from three to 18, with higher scores indicating greater suicide risk. A total score of seven or more is considered at risk for suicide17. The SBQ-R had acceptable reliability in the current sample (Ī±ā€‰=ā€‰0.795)15.

Demographic determinants

The demographic determinants assessed were self-reported and included the following: age (in years), gender, country of birth, ethnicity, Australian residency or visa status, history of mental health problems, and presence of a disability. Presence of disability was defined as the presence of long-term physical, mental, intellectual, or sensory impairments, which may hinder full and effective participation in society.

Occupational determinants

The occupational determinants assessed included both PhD and non-PhD related occupational factors: PhD study load (full-time, part-time), PhD funding (fees and stipend, fees only and self-funded), year of study, study hours (average reported weekly hours spent on PhD study), work hours (average reported weekly hours in employment), and total work and study hours (total reported weekly hours spent on work and study).

Psychological determinants

Imposter thoughts

The 20-item Clance Impostor Phenomenon Scale (CIPS)18 was used to measure imposter thoughts. In the CIPS, respondents are asked to indicate the extent to which each statement (e.g., ā€œI can give the impression that Iā€™m more competent than I really am.ā€) is true on a 5-point scale (1 ā€˜Not at all trueā€™ to 5 ā€˜Very trueā€™). Scores on each statement are summed. The total score can range from 20 to 100, with higher scores indicating greater frequency and seriousness of imposter thoughts. The scale had excellent reliability in the current sample (Ī±ā€‰=ā€‰0.944)15.

Perfectionism

The 8-item Short Almost Perfect Scale (SAPS)19 was used to measure perfectionism. The SAPS has two subscales (SAPS-standards, SAPS-discrepancy) that each contain four items. The SAPS-standards measures perfectionistic standards (e.g., ā€œI have high standards of myself.ā€) and the SAPS-discrepancy measures discrepancy between standards and perceived performance (e.g., ā€œI am hardly ever satisfied with my performance.ā€). In the SAPS, respondents are asked to indicate the extent they agree with each statement (1 ā€˜Strongly disagreeā€™ to 7 ā€˜Strongly agreeā€™) on a 7-point scale. Scores on each statement are summed. The total score for each subscale can range from four to 28, with higher scores indicating greater perfectionism standards or discrepancy. The SAPS-standard and SAPS-discrepancy subscales had good and excellent reliability in the current sample (Ī±ā€‰=ā€‰0.869 and Ī±ā€‰=ā€‰0.908, respectively)15.

Social determinants

Loneliness

The 20-item University of California, Los Angeles (UCLA) Loneliness Scale20 was used to measure loneliness. In this scale, respondents are asked to indicate the frequency to which they experience feelings of loneliness (e.g., ā€œI have nobody to talk to.ā€) on a 4-point scale (3 ā€˜I often feel this wayā€™ to 0 ā€˜I never feel this wayā€™). Scores on each item are summed. The total score can range from zero to 60, with higher scores indicating greater experiences of loneliness. The scale had excellent reliability in the current sample (Ī±ā€‰=ā€‰0.958)15.

Social group membership

The 4-item Multiple Group Membership scale (MGM)21 was used to measure social group membership. In the MGM, respondents are asked to indicate the extent they agree with each item (e.g., ā€œI have strong ties with lots of different groups.ā€) on a 7-point scale (1 ā€˜Strongly disagreeā€™ to 7 ā€˜Strongly agreeā€™). Scores on each item are summed. The total score can range from four to 24, with higher scores indicating greater social group belonging. The scale had excellent reliability in the current sample (Ī±ā€‰=ā€‰0.905)15.

Relational determinant

Supervisorā€“student relationship

The Questionnaire on Supervisor-Doctoral Student Interaction (QSDI)22 was used to assess the nature of the supervisorā€“student relationship. The QDSI measures eight types of supervisory behaviour (eight octants), which are then represented as two dimensions of the supervisory-student relationship: agency (QSDI-A; i.e., influence and leadership) and communion (QSDI-C; i.e., proximity and cooperativeness). In the QSDI, respondents are asked to indicate the frequency to which they experience the behaviours of their supervisors (e.g., ā€œMy supervisor acts irritable with me.ā€). Instructions and details on the scoring of the QSDI can be requested from Mainhard et al.22. Higher scores on the QSDI-A and QSDI-C indicate a greater degree of supervisory agency and communion, respectively. The eight octants had (at minimum) acceptable levels of reliability in the current sample (Ī±ā€‰=ā€‰0.731ā€“0.889)15.

Procedure

Participation took place online via a survey programmed in Qualtrics. Individuals interested in participating in the research study were asked to read the information sheet, and then provide their informed consent for participation. Following this, participants responded to questions relating to their demographics, mental health, and occupational, psychological, social, and relational determinants of mental health. All responses in the survey were optional.

Analysis

Analyses were conducted in SPSS version 27.0 for Windows. Visual inspection of the data suggested that scale data were normally distributed, and there were no extreme outliers. Although some scales demonstrated mild skew and kurtosis, the use of a robust analysis (regression) and the sample size were considered sufficient to overcome these variations in the data23.

Associations between mental health measures and determinants (predictor variables) were assessed using independent samples t-tests, one-way ANOVA test and bivariate correlations. Linear hierarchical regression analysis was used to test significant predictor variables for each of the three mental health measures. Pairwise deletion was used to handle missing data, such that if the participant did not respond to one or more items in a measure, their data for that measure was treated as missing and was not included in the analysis24. The block structure for each of the three regression models was based on the nature of the variables and their susceptibility to change. Therefore, the demographic variables, which are less susceptible to change, were entered into the first block, and then the occupational, psychological, social, and relational variables, which are more susceptible to change, entered into the later blocks8. The block structure allows us to examine the contribution of the different types of variables as a group and as individual variables, and draw conclusions about the types of predictors that contribute most to the mental health of PhD students.

For the linear hierarchical regression analysis, some data was binarized due to small sample sizes in minority groups. For example, very few participants identified as gender diverse and therefore were grouped with female participants, and two participants were missing data. The decision was premised on bivariate results that indicated the data provided by the two groups were the least different (and not significantly different) from each other. Additionally, PhD funding (fees and stipend, fees only, self-funded) was binarized to whether participants received stipend funding (fees and stipend) or not (fees only or self-funded). The distribution of residuals appeared normal for all regression models. Collinearity diagnostics were good, with Variance Inflation Factor (VIF) values all less than 2.5, indicating no influential collinearity25.

Results

Sample characteristics

The sample characteristics are summarised in Table 1. The mean age of the sample was 34.12Ā years (SDā€‰=ā€‰10.32). In this sample, 219 participants self-reported to be female, 73 male, and 8 gender diverse. The majority of the sample was born in Australia (67.55%) and held an Australian citizenship (78.48%). A large proportion of the sample had a history of mental health problems and had received a formal diagnosis (45.70%). Most participants reported that they did not have a disability (76.49%). The majority of the sample were enrolled in their PhD program in a full-time capacity (80.79%).

Table 1 Sample characteristics.
Full size table

The descriptive statistics of measures are summarised in Table 2. In brief, 45.36% of the sample were considered to have met the threshold for moderate to severe depression, 39.40% for moderate to severe anxiety, and 36.42% were considered at risk for suicide.

Table 2 Descriptive statistics of the measures implemented in the study.
Full size table

Univariate analysis

Relationships between scores on the mental health measures and categorical predictor variables are presented in Table 3. Levels of depression, anxiety, and suicidality were significantly higher in participants who did not identify as male (psā€‰ā‰¤ā€‰0.018), those with a history of mental health problems (psā€‰<ā€‰0.001), and a disability (psā€‰ā‰¤ā€‰0.024). Levels of depression and anxiety were significantly lower in participants who had partial funding for their PhD program (ā€˜fees onlyā€™ funding) compared to those who had full funding (ā€˜fees and stipendā€™ funding) or no funding (self-funded) (psā€‰ā‰¤ā€‰0.036). Levels of suicidality were significantly higher in participants who were Australian citizens compared to those who were not (pā€‰<ā€‰0.001). No other relationships were significant.

Table 3 Univariate associations between mental health measures and categorical predictor variables.
Full size table

Bivariate analysis

Relationships between scores on the mental health measures and continuous predictor variables are presented in Table 4. Higher levels of depression, anxiety, and suicidality were significantly correlated with greater frequency and seriousness of imposter thoughts (psā€‰<ā€‰0.001), greater perfectionism discrepancy (psā€‰<ā€‰0.001), greater experiences of loneliness (psā€‰<ā€‰0.001) and lower supervisor-student communion (psā€‰ā‰¤ā€‰0.021). Higher levels of depression symptoms were significantly correlated with lower age (pā€‰<ā€‰0.001), greater perfectionism standards (pā€‰<ā€‰0.047), and lower levels of group belonging (pā€‰=ā€‰0.017). Higher levels of anxiety symptoms were significantly correlated with lower age (pā€‰<ā€‰0.001), greater total work and study hours (pā€‰=ā€‰0.047), and greater perfectionism standards (pā€‰<ā€‰0.001). Finally, higher levels of suicidality were significantly correlated with lower levels of group belonging (pā€‰<ā€‰0.001). No other relationships were significant.

Table 4 Bivariate associations between mental health symptom measures and continuous predictor variables.
Full size table

Hierarchical regression

Significant univariate and bivariate associations (see Tables 3 and 4) were added to the hierarchical regression models in blocks by predictor category. Overall, predictor variables in each regression model explained a significant amount of variance for depression (52.2%), anxiety (49.4%), and suicidality (27.3%). All blocks were significant, except for the occupational block in the depression model and the relational block in the suicidality model. Results from the hierarchical regression analysis are presented in Table 5.

Table 5 Results from Hierarchical Regression Analysis.
Full size table

In the depression model, demographic variables such as lower age (Ī²ā€‰=ā€‰āˆ’Ā 0.150, pā€‰=ā€‰0.008), having history of mental health problems (Ī²ā€‰=ā€‰0.285, pā€‰<ā€‰0.001), and having a disability (Ī²ā€‰=ā€‰0.119, pā€‰=ā€‰0.037) predicted depression symptoms. Psychological variables such as greater frequency and seriousness of imposter thoughts (Ī²ā€‰=ā€‰0.326, pā€‰<ā€‰0.001) and greater perfectionism discrepancy (Ī²ā€‰=ā€‰0.294, pā€‰<ā€‰0.001) predicted depression symptoms. Social variables such as greater experiences of loneliness (Ī²ā€‰=ā€‰0.300, pā€‰<ā€‰0.001), and relational variables such as lower student-supervisor communion (Ī²ā€‰=ā€‰āˆ’Ā 0.266, pā€‰<ā€‰0.001) also predicted depression symptoms.

In the anxiety model, demographic variables such as lower age (Ī²ā€‰=ā€‰āˆ’Ā 0.118, pā€‰=ā€‰0.039) and a history of mental health problems (Ī²ā€‰=ā€‰0.275, pā€‰<ā€‰0.001) predicted anxiety symptoms. Occupational variables such as greater hours spent on work and study significantly predicted anxiety symptoms (Ī²ā€‰=ā€‰0.161, pā€‰=ā€‰0.004). Psychological variables, including greater frequency and seriousness of imposter thoughts (Ī²ā€‰=ā€‰0.272, pā€‰<ā€‰0.001) and greater perfectionism discrepancy (Ī²ā€‰=ā€‰0.296, pā€‰<ā€‰0.001), predicted anxiety symptoms. Social variables such as greater experiences of loneliness (Ī²ā€‰=ā€‰0.282, pā€‰<ā€‰0.001) and relational variables such as decreased student-supervisor communion (Ī²ā€‰=ā€‰āˆ’Ā 0.220, pā€‰<ā€‰0.001) also predicted anxiety symptoms.

In the suicidality model, demographic variables such as having a history of mental health problems (Ī²ā€‰=ā€‰0.291, pā€‰<ā€‰0.001) and having a disability (Ī²ā€‰=ā€‰0.126, pā€‰=ā€‰0.026) predicted suicidality. Psychological variables, including greater frequency and seriousness of imposter thoughts (Ī²ā€‰=ā€‰0.157, pā€‰=ā€‰0.030) and greater perfectionism discrepancy (Ī²ā€‰=ā€‰0.149, pā€‰=ā€‰0.031), and the social variable, greater experiences of loneliness (Ī²ā€‰=ā€‰0.312, pā€‰<ā€‰0.001), were also significant predictors of suicidality.

Discussion

Findings from the present study paint a bleak picture of the mental health of PhD students in Australia. In our sample of PhD students, 45% met symptom thresholds of moderate to severe depression, 39% met symptom thresholds of moderate to severe anxiety, and 36% were considered at risk of suicide. These figures are comparable to those observed in PhD students in the UK3,8 and appear higher than those observed in the Australian general community26,27. Importantly, the mental health symptoms experienced by PhD students in our sample were predicted by a set of demographic, occupational, psychological, social, and relational factors. We discuss our findings from these determinants in the order of predictive ability.

The psychological predictors, as a group (or block), were strongest in predicting depression and anxiety in our sample. Frequent imposter thoughts were experienced by a large proportion of our sample (42%), with over 26% experiencing serious, intense imposter thoughts. The frequency and seriousness of imposter thoughts predicted symptoms of depression, anxiety, and suicidality (see also Berry et al.8) and were associated with perfectionism standards (and discrepancy). However, only perfectionism discrepancy (and not standards) predicted mental health symptom outcomes. Perfectionism standards may be considered an adaptive behaviour to reduce stress and anxiety, by holding oneself to a higher standard28. This behaviour can become maladaptive when these standards are not met (i.e., perfectionism discrepancy). Previous studies have demonstrated that maladaptive perfectionists have poorer psychological health, life satisfaction, and self-esteem, when compared to adaptive perfectionists29.

Social determinants were also important predictors of mental health symptom outcomes in our sample. Of the social determinants examined, loneliness was the strongest and most consistent individual predictor across the three mental health outcomes: depression, anxiety, and suicidality (see also Berry et al.8). PhD students can experience loneliness in many ways. For example, some PhD students may feel lonely in a small academic department with few other students while others may feel lonely in a large academic department if there are fewer shared interests (intellectual or non-intellectual interests). Despite loneliness being an important predictor of mental health outcomes, experiences of social group membership were not. This may be due to a potential overlap in the predictors of loneliness and social group membership in our study. This would not be surprising, as one could argue that the underlying protective mechanism of social group membership is the feeling of being socially connected, rather than belonging to different groups30. An important consideration here is the impact of COVID-19, which saw a rise in online learning and study models31. While one could argue that online means allow for more social group belonging, individuals could still experience loneliness (ā€œan illusion of companionship without the demands of friendshipā€32).

The relational determinant, student-supervisor relationship, was also an important predictor for mental health symptom outcomes, in particular, student-supervisor communion, or the proximity and cooperativeness of the supervisor. The importance of the student-supervisor relationship on mental health outcomes was also observed by Berry et al.8. However, in their study, they found that student-supervisor communion and agency (i.e., influence and leadership) predicted mental health outcomes. The discrepancy between studies may indicate differences in perceived PhD student supervision quality in the UK and Australia. Without further investigation into the differences in supervision quality perceived by the two PhD student populations, it is difficult to pinpoint the reason for the differences in results. Nevertheless, findings from our study emphasise the importance of supervision that focuses on collaboration and flexibility. It is possible that the provision of guidance during the PhD study (agency) is important to student progress, but not to mental health outcomes. Rather, it is how this guidance is provided (communion) that affects their mental health. For example, a supervisor can provide guidance with no room for the student to negotiate (high agency, low communion) or guidance that either adapts to the studentā€™s way of working or allows for negotiation (high agency, high communion). In this sense, the communion aspect of the student-supervisor relationship is likely to impact the mental health of PhD students.

Demographic and occupational determinants such as age, history of mental health problems, and total work and study hours predicted mental health symptom outcomes. Compared to other determinants (i.e., social, psychological, relational) studied, the effects of demographic and occupational determinants were smaller. This can be viewed positively, as these factors are stable and less susceptible to change. Notably, unlike previous studies33,34, the number of years of PhD study did not predict mental health symptom outcomes. It is possible that the negative impact on mental health arises not from the number of years of study per se, but instead, the perceived progress of the PhD study, which may be of interest to future research.

Limitations and future directions

The findings from the present study must be interpreted within the context of some limitations. First, the study design was cross-sectional, and therefore, while findings from the study explain associations between the determinants and the mental health outcomes, it is limited in its ability to establish causation or any long-term effects of the determinants of the mental health of PhD students. Future research should determine the causal role of these determinants using longitudinal research models.

Second, the study investigated a wide range of determinants of mental health, but there may be other important determinants such as socio-economic status, hopefulness, or perceived PhD progress. Future research should explore additional determinants that may influence the mental health of PhD students.

Third, the voluntary recruitment of research participants may result in a bias in the study sample. For example, PhD students who have experienced poor mental health may have been more likely to participate in the study than those who have not, which could overestimate the proportion of PhD students experiencing mental health problems in this population. Future research should explore other ways of recruitment to reduce this sampling bias. For example, implementing mental health surveys and monitoring as part of progress reviews undertaken by PhD students.

Conclusion

In conclusion, the present study demonstrated several determinants of depression, anxiety, and suicidality in PhD students in Australia. More specifically, demographic determinants (age, history of mental health problems, presence of disability), occupational determinants (total hours spent on work and study), psychological determinants (imposter thoughts, perfectionism discrepancy), social determinants (loneliness, social group memberships), and relational determinants (student-supervisor relationship communion). Among these determinants, the social determinant, loneliness, was the most consistent individual predictor of the mental health of PhD students across the three mental health outcomes examined. The unique challenges of completing a PhD program should be recognised and strategies must be implemented to protect the mental health of this student population. Based on findings from the present study, universities are encouraged to explore ways to mitigate experiences of loneliness in PhD students, promote psychological supports, and to foster effective, collaborative relationships between students and their supervisors.